In our experience in non-biomedical domains there seem to be two types of applications. In the first, one can use what may be called a ``noun-driven'' approach. The type of an entity is highly predictive of its role in the event. In this case, it is not so necessary to get the Subject-Verb-Object relations correct. Looser patterns can be written. For example, if the only patterns we are looking for are
Protein inhibits Reaction
Protein promotes Reaction
Then the protein always fills the role of the effector and the reaction always fills the role of the effected.
In other domains, the roles of entities in events cannot be predicted from their type, but only from their syntactic place in sentences. These applications require what may be called a ``verb-driven'' approach. Tighter patterns must be written, and Subject-Verb-Object relations must be discovered. For example, in
Protein binds to Protein
we cannot tell from the fact that something is a protein which of the two roles it plays in the binding event.
The vast specialized and highly organized terminology of biomedicine suggests that perhaps a noun-driven approach would be adequate. The roles of entities may be very tightly constrained. On the other hand, as Friedman et al. (2001) have shown, there can be deeply nested relations in complex events, and it can be crucial to get the Subject-Verb-Object relations right, in which case a verb-driven approach is required.